Grant: Discovery Grants Program - Individual

Suppose you try to remember a list of words in which a digit was inserted. You are more likely to remember the digit than the words. This von Restorff effect nicely illustrates the powerful influence of distinctiveness (Hunt, 2013), one of the few universal memory principles (Surprenant & Neath, 2009). Accordingly, many memory models include distinctiveness at their heart (see, Neath & Brown, 2007, for a review). Among them, the Computational Feature Model we proposed stands out because it can apply equally well to short-term and long-term memory and to tasks such as free recall, order reconstruction, immediate serial recall and item recognition (e.g., Cyr et al., 2022). The current proposal is aimed at developing an improved version of the Computational Feature Model that will systematically be tested through multiple series of experiments organised in four streams. In the Computational Feature Model, items are represented as vectors of features. Item presentation automatically generates a vector in primary memory and an identical vector in secondary memory. Traces in primary memory are subject to similarity-based retroactive interference, but overwritten features can be restored by a rehearsal process. At recall, degraded traces in primary memory are used as cues to enable the selection of a retrieval candidate from secondary memory. In the first stream, we will perform a comprehensive test of the innovations we introduced to the model (Saint-Aubin et al., 2021) through a series of pre-registered experiments based on a priori predictions. More specifically, using the production effect, that is the better recall of items read aloud compared to silent items, we will test the details of the proposed interference and rehearsal processes. In addition, the capacity of the Computational Feature Model to account for output order in free recall, which is a major strength of the model (see, Murdock, 2008), will be systematically tested. In the second stream, new empirical effects predicted by the Computational Feature Model will be investigated. Furthermore, we will test the capacity of the Computational Feature Model to account for conflicting findings across memory tasks. In the third stream, the Computational Feature Model will be used to test hypotheses explaining the beneficial impact of semantic similarity on item recall, as the latter poses a challenge to all similarity-based models. Finally, in the fourth stream, we will move the Computational Feature Model from a trial-unit model to a continuum model with no built-in boundaries between successive trials. Among other things, this could allow the model to account for inter-trial interference effects such as those associated to proactive interference. The experimental database generated by the predictions derived from the model will establish new benchmark effects for testing memory models.